LH
L.C.A. Huijbregts
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Ultra-low power Edge AI hardware is in increasing demand due to the battery-limited energy budget of typical Edge devices such as smartphones, wearables, and IoT sensor systems. For this purpose, this Thesis introduces an ultra-low power event-driven SRAM-based Compute In-Memory (CIM) accelerator optimized for inference of Binary Spiking Neural Networks (B-SNNs). In this Thesis, a custom-designed 3nm SRAM cell is developed, with up to four read ports to improve inference performance and one transposable read/write port for efficient on-chip learning functionality. The event-based nature of SNNs is exploited to minimize the computation and memory cost. The design benefits from technology scaling of fully digital design by synthesizing the accelerator in the imec 3nm FinFET technology node. The proposed accelerator's performance is evaluated by running MNIST inference at 97.6% accuracy, achieving an impressive throughput of 44M inferences/s at 607 pJ/inference (3.2 fJ per synaptic operation) while running at 29 mW. The results demonstrate that the proposed accelerator provides an energy-efficient and high-performance solution for inference of Binary SNNs, opening up new possibilities for Edge AI applications.
...
Ultra-low power Edge AI hardware is in increasing demand due to the battery-limited energy budget of typical Edge devices such as smartphones, wearables, and IoT sensor systems. For this purpose, this Thesis introduces an ultra-low power event-driven SRAM-based Compute In-Memory (CIM) accelerator optimized for inference of Binary Spiking Neural Networks (B-SNNs). In this Thesis, a custom-designed 3nm SRAM cell is developed, with up to four read ports to improve inference performance and one transposable read/write port for efficient on-chip learning functionality. The event-based nature of SNNs is exploited to minimize the computation and memory cost. The design benefits from technology scaling of fully digital design by synthesizing the accelerator in the imec 3nm FinFET technology node. The proposed accelerator's performance is evaluated by running MNIST inference at 97.6% accuracy, achieving an impressive throughput of 44M inferences/s at 607 pJ/inference (3.2 fJ per synaptic operation) while running at 29 mW. The results demonstrate that the proposed accelerator provides an energy-efficient and high-performance solution for inference of Binary SNNs, opening up new possibilities for Edge AI applications.
Bachelor thesis
(2020)
-
L.C.A. Huijbregts, M.A. Jongepier, J.A. Martinez Castaneda, R.C. Hendriks, J.E.J. Schmitz, S. Izadkhast
Public Address systems that include a setup with at least one microphone and speaker can suffer from acoustic feedback. This results in an annoying howling effect which can damage hardware and human hearing. To solve this issue, an adaptive filter that estimates the feedback path and uses this estimate to cancel the feedback can be designed. However, because the adaptive filter receives signals from both the microphone and the feedback, and because sound signals are generally correlated over time, the estimate becomes biased. To reduce this bias, the speaker signal can be decorrelated from the input. In this thesis several options to decorrelate these signals are explored, and they are evaluated based on decorrelation performance and effect on audio quality. Frequency shifting is selected as the best decorrelation method as it provides the most decorrelation while retaining audio quality. Finally it is shown that using Frequency Shifting to decorrelate the microphone and speaker signal indeed improves the estimation of the feedback path.
...
Public Address systems that include a setup with at least one microphone and speaker can suffer from acoustic feedback. This results in an annoying howling effect which can damage hardware and human hearing. To solve this issue, an adaptive filter that estimates the feedback path and uses this estimate to cancel the feedback can be designed. However, because the adaptive filter receives signals from both the microphone and the feedback, and because sound signals are generally correlated over time, the estimate becomes biased. To reduce this bias, the speaker signal can be decorrelated from the input. In this thesis several options to decorrelate these signals are explored, and they are evaluated based on decorrelation performance and effect on audio quality. Frequency shifting is selected as the best decorrelation method as it provides the most decorrelation while retaining audio quality. Finally it is shown that using Frequency Shifting to decorrelate the microphone and speaker signal indeed improves the estimation of the feedback path.